llms.txt generator. Tell AI tools what your site is.

What is llms.txt?

llms.txt is a proposed standard from 2024 (llmstxt.org): a plain text file at the root of your website, written in Markdown, that tells AI tools what the site is and where its important pages are. Think of it as a map written for a language model: shorter and cleaner than your pages, without the menus, banners and scripts around the text.

It lives at one fixed address, yourdomain.dk/llms.txt, the way robots.txt does. But where robots.txt says what crawlers may read, llms.txt says what is worth reading.

What to put in your llms.txt

The format is short, and always in this order:

• A title, # and the name of your business or website. The only line the standard requires.
• A summary on a line starting with >: one or two sentences on what you do, for whom and where.
• A few lines of details, if they help: your delivery area, how bookings work, what you don’t do.
• Sections, each a ## heading over a list of links. Each link is the page’s name in square brackets, its address in round brackets, and a note on what the page holds.
• A section called “Optional”, for pages an AI tool can skip when it is short on space.

Link the pages a customer would ask about: services, prices, opening hours, contact and booking. Leave out the login page, the cookie policy and anything that isn’t public. Ten good links beat a hundred.

What is ai-catalog.json?

ai-catalog.json belongs to Agentic Resource Discovery (ARD), a newer proposal for how AI agents find the agents, MCP servers and APIs a website offers. It lives at /.well-known/ai-catalog.json and lists them as entries.

Most small business websites offer none of those, and that is fine. The generator makes a catalog with your name, a link to your llms.txt and an empty list of entries: valid, honest, and it passes Lighthouse’s check. Without the file, Lighthouse normally reports the check as not applicable rather than failed.

How to add llms.txt to your website

1. Fill in your name, your website’s address and a short summary.
2. Add your most important pages, grouped in sections. A path like /menu is fine; it becomes a full address.
3. Check that the three Lighthouse checks beside the result are ticked.
4. Download llms.txt and upload it to the root of your website, so it opens at https://yourdomain.dk/llms.txt.
5. Download ai-catalog.json and upload it to a folder called .well-known, so it opens at https://yourdomain.dk/.well-known/ai-catalog.json.
6. Open both addresses in your browser, then run your site through PageSpeed Insights and look under Agentic Browsing.

Not every website builder lets you put files at the root of your own domain. If yours doesn’t, ask its support, or look for an SEO or AI setting that makes the file for you.

Does llms.txt help? An honest answer

It helps the AI tools that read it: an assistant asked about your business, or an agent sent to book a table, gets a clean map instead of guessing from your menus. And in 2026 Google’s Lighthouse, the engine behind PageSpeed Insights, began checking llms.txt and ai-catalog.json in a new Agentic Browsing category.

What it does not do is rank you higher. No major search engine or AI service has said it uses llms.txt for ranking, and Google has said its search doesn’t need it for AI features. Treat it like a sitemap for AI: quick to make, worth having, no miracle. Useful pages about what you do, on a fast website, still matter most.

What Lighthouse checks

PageSpeed Insights shows both checks under Agentic Browsing, in the Agent Discoverability group:

• llms.txt: that it has a # title, at least one link written in Markdown, and more than a few words.
• llms.txt: that the address answers. A missing file is “not applicable”; a server error fails.
• ai-catalog.json: that it is valid JSON with specVersion "1.0", an entries list, and only the fields the ARD schema defines.
• ai-catalog.json: that every entry has an identifier starting urn:air:, a displayName, a type, and either a url or the data itself.

Common mistakes

• Links as bare addresses. Lighthouse only counts links written in Markdown, so a list of plain URLs fails.
• A web page instead of a text file. Some website builders answer every unknown address with a page and status 200, so Lighthouse reads HTML and finds no title.
• The file in the wrong place: llms.txt belongs at the root, not in a subfolder or on a subdomain.
• Extra fields in ai-catalog.json, such as a name or description at the top. The schema allows only specVersion, host and entries there.
• Links to pages that are gone, private or hidden from search. Update the file when your pages change.